Cups Props and Vanes

Risø–R–766(EN)
Cups, Props and Vanes

Leif Kristensen
Risø National Laboratory, Roskilde, Denmark

August 1994
Abstract We discuss the dynamics of the cup anemometer, the propeller ane-
mometer and the wind vane. The phenomenological model by Kristensen (1993)
is modified to describe the motion of the propeller anemometer. We use the well-
know second-order differential equation describing the motion of the wind vane
to study the properties of the vane itself and the vane in combination with a cup
and a vane. It is argued that even the simplest question about how to record the
signal from a cup anemometer has an ambiguous answer and can lead to a, not
necessarily small, systematic error in the measured mean wind speed. We show
how it is possible to measure the entire wind direction variance by utilizing the
fact that an underdamped vane by ‘overshooting’ may compensate for the high-
frequency variance loss if the damping coefficient is about 0.4. Finally we discuss
possible sources of bias on the measured mean-wind speed when using a propeller-
vane anemometer. It turns out that this anemometer has the same type biases from
the lateral and vertical wind fluctuations as has the cup anemometer. In addition
there are bias contributions from misalignment between the mean wind and the
propeller axis and from the translatory motion of the propeller itself with respect
to the vertical wind-vane axis.
ISBN 87–550–2006–2
ISSN 0106–2840
Information Service Department · Risø · 1994

Contents
1 Introduction 5
2 The Cup Anemometer 5

2.1 Phenomenological Model 6
2.2 Anemometer Dynamics 7
2.3 Mean Wind-Speed Measurements 11
3 The Propeller Anemometer 13
4 The Wind Vane 14

4.1 Wind-Vane Dynamics 15
4.2 Measuring the Lateral Velocity Variance 19
5 Cup and Vane 26
6 Prop and Vane 27
7 Conclusions 34
References 37
Risø–R–766(EN) 3
1 Introduction
Probably the most common anemometers are the rotating anemometers, cups
and propellers, and the Pitot tube. They are often used together with a wind vane
for wind-direction determination. Such a combination is a sturdy and reliable
instrument package for the determination of the mean of the horizontal wind-
velocity component. It is easy to operate and is used in weather stations, airports,
wind farms and sites where large structures such as bridges are under construction.
(The Pitot tube without wind vane is standard instrumentation in aircraft for
measuring airspeed.)
As indicated in the title, we are here concerned only with the rotating anemometers
and the wind vane. There is a vast amount of literature about these instruments.
Wyngaard (1981) has given a very well written summary and also provided a com-
prehensive list of useful references. There are two major reasons why so much has
been written about the cup anemometer, which was invented as early as 1846 by
the Irish astronomer Thomas Romney Robinson (Middleton, 1969 and Wyngaard,
1981) and which, in almost the same standard design, is still in use. One is the
remarkable linearity of the calibration, the other the so-called overspeeding which,
allegedly, should cause large systematic errors when trying determine of the mean
wind speed in more than light turbulence. Kristensen (1993) discussed this over-
speeding in much detail and developed a phenomenological model for the forcing
of the cup-anemometer rotor. Here we limit ourselves to a short description of this
model and the experimental justification for it and to a discussion of the proper
signal processing with and without a wind-vane signal.
It is postulated that the forcing model for the cup anemometer with a slight
modification can be applied to a propeller. The vane alone and the propeller vane
will be discussed in great detail in order to specify, as in the case of the cup
anemometer, what kind of systematic errors must be expected when measuring
the mean wind speed.
2 The Cup Anemometer

Figure 1 shows a cup anemometer with three cups. As a field instrument and
general-purpose anemometer for operational purposes the cup anemometer has at
least two good properties. It is omnidirectional; when mounting the instrument,
it is only necessary to make sure that the axis is pointing in the vertical direction.
Further, the cup anemometer is robust and easy to operate.
We assume that the cup anemometer is mounted with its axis vertical. In this case
the equation for the rotation rate of the rotor can be written

s̃˙ = F (s̃, ũ2 + ṽ 2 , w̃), (1)
where ũ, ṽ and w̃ are the instantaneous horizontal and vertical wind components
and s̃ is the instantaneous rotation rate of the anemometer rotor in rad s−1 . The
angular momentum of the rotor is proportional to s̃ and (1) just states that the
rate of change of the angular momentum is equal to the torque on the rotor. This
torque is caused by the wind and the friction in the instrument bearings. For a
Figure 1. The Risø cup anemometer. The height is 0.26 m.
real cup anemometer the right-hand side, the torque divided by the√moment of
inertia, is a function of s̃, the total horizontal wind component h̃ = ũ2 + ṽ 2 as
well as the vertical wind component w̃.
2.1 Phenomenological Model

It was shown by Kristensen (1993) that we can explain behaviour of the cup
anemometer by assuming that the function F has the form
(h̃ + µ1 w̃ − s̃)(h̃ + µ1 w̃ + Λs̃) µ2 w̃2

F (s̃, h̃, w̃) = + . (2)
0 ( + Λ) 20
Here , 0 , Λ, µ1 and µ2 are instrument constants, the first three with the dimension
of length and the last two dimensionless. The first quantity is the calibration
distance which, under conditions with constant magnitude U of the horizontal wind
velocity, can be interpreted as the length of the column of air which has to pass
through the anemometer for the cup rotor to turn one radian. The second quantity
0 is the so-called distance constant which is the length of the column of air which
has to pass through the anemometer before it has reacted with 1 − exp(−1) ≈ 0.63
of its final response to a sudden change in the wind speed. The last instrument
length scale Λ cannot be interpreted as directly as and 0 . Finally, the constants
µ1 and µ2 characterize the angular response of the cup anemometer up to second
order in the angle between the direction of the wind vector and the rotor plane;
if they are both zero this response is a cosine response, which means that the
anemometer is insensitive to the vertical wind component. It is interesting that if
we can construct an anemometer with Λ = , µ1 = 0 and µ2 = 1 this instrument
6 Risø–R–766(EN)
will be responding only to the magnitude of the three-dimensional velocity vector
since
ũ2 + ṽ 2 + w̃2 − 2 s̃2

F (s̃, h̃, w̃) = . (3)
20
Equations (1) and (2) imply that the steady-state calibration of the cup anemom-
eter is linear without offset. This is easily seen by setting s̃˙ and w̃ equal to zero
and solving for S = s̃. The result is
U
S= . (4)

Experience actually shows that the calibration is indeed approximately linear with
an offset U0 which in most cases can be ignored when the wind speed is more than
a few meters per second∗ . Brazier (1914) and Patterson (1926) demonstrated that
the linearity is better the shorter the diameter of the cup rotor. Patterson (1926)
determined, for a number of cup anemometers, the anemometer factor, defined as
U
f= = , (5)
rS r
where r is the cup radius, i.e. the distance from the axis to the center of one
of the cups. He found that, depending on cup diameter and r, the anemometer
factor varies between 2.5 and 3.5. We expect and Λ to be dependent on only
the anemometer geometry and independent of the mass and mass distribution of
the cup rotor. For the Risø cup anemometer shown in Figure 1, = 20 cm and
r = 6 cm. Correspondingly, f = 3.3.
The distance constant depends on the moment of inertia I, i.e. of the cup-rotor
mass distribution, and is given by
2I
0 = , (6)
ρCAr( + Λ)
where C is a dimensionless constant of order unity, ρ ≈ 0.0013 g cm−3 the density

of air and A an effective cup area.
Since I is proportional to the rotor density ρr and to the fifth power of its linear
dimensions (∝ r) we conclude from (6) that for geometrically similar rotors (A ∝
r2 ) the distance constant is proportional to ρr /ρ and to r. The rotor of the Risø
model (Busch et al., 1980) in Figure 1 is made of carbon reinforced plastic with a
density ρr of about 1.5 g cm−3 . The distance constant 0 was determined in wind-
tunnel studies to be 170 cm. This is a rather modern and sturdy cup anemometer,
which is used for routine measurements of mean wind speed by Risø National
Laboratory. Older models are typically made of steel and they are often larger,
with radii of about 15 cm. They have distance constants of about 20 m and reacts
consequently much slower than newer models of standard cup anemometers.
2.2 Anemometer Dynamics

When exposed to the fluctuating, turbulent wind the first-order response of a cup
anemometer is that of a first-order filter. Assuming that the mean of the vertical
∗ If the offset must be taken into account this is done by replacing, in (2), h̃ by h̃ − U . Then
0
the calibration expression (4) becomes S = (U − U0 )/.
velocity component is zero and that the coordinate system is chosen such that
the lateral velocity component ṽ also has a zero mean, we may decompose the
instantaneous wind vector and the anemometer response according to
   

 
 
 


 ũ 
 
 U +u 


 
 
 


 
 
 


 
 
 

 ṽ 
  
 v 

= . (7)

 
 
 


 w̃ 
 
 w 


 
 
 


 
 
 


 
 
 

 s̃ 
  
 S+s 

Kristensen (1993) has shown that first-order perturbation applied to (1) with F
given by (2) leads to
s 1 u w
ṡ + = + µ1 , (8)
τ0 τ0 τ0
where
0
τ0 = . (9)
U
Equation (8) is the equation for a first-order, linear filter with the time constant
τ0 which, as (9) shows, is inversely proportional to the magnitude of mean-wind
velocity.
The properties of this filter can by studied in a number of ways. One simple and
illuminating diagnostic tool is a study of its response to a step function given by



 0 for t < 0
u(t) = (10)


 ∆U for t ≥ 0
with w(t) = w̃ kept equal to zero.

In this case, the response will be
∆U
s(t) = (1 − e−t/τ0 ) (11)

—the classical response of a first-order system to a step function. Figure 2 shows

how the response adjusts to the new equilibrium value of the input. It grows such
that when t = τ0 it has attained ≈ 63% of its terminal value s(∞).
Applying Taylor’s hypothesis (for ‘frozen turbulence’), we may convert the tem-
poral variable to a spatial variable in the direction x of the mean wind by
x=U ×t (12)
in which case (8) takes the form
1
ω 0.5
∆U
0
0 1 2 3
t/τ0
Figure 2. First-order response (11) of a cup anemometer to a step input ∆U .
ds s 1 u w
+ = + µ1 . (13)
dx 0 0 0
In (13) the coefficients are all instrument constants, independent of the wind speed
U . The implication is that the cup anemometer is a spatial rather than an temporal
filter. This is fortunate because atmospheric turbulence in most cases, i.e. when
the mean wind is large compared to the r.m.s. fluctuations of the turbulence, can
be considered a random, ‘frozen’ velocity field which is being transported through
the observation point by the mean wind.
The first-order perturbation equation (13) is sufficient for applications where we
are concerned with spectral filtering of the velocity signal. However, if we want
to quantify a phenomenon as overspeeding we will need to derive a second-order
perturbation equation, as shown in Kristensen (1993).
Following Wyngaard et al. (1974), we write this equation in the dimensionless
form
d s s u w
0 + = a1 + a2
dx S S U U
s 2 u 2 w 2
+ a3 + a4 + a5
S U U
s u u w w s
+ a6 + a7 + a8 . (14)
S U U U U U
Wyngaard et al. (1974) now determined experimentally the eight coefficients

a1 , . . . , a8 in a wind tunnel for a particular cup anemometer. They did that by
measuring the torque I × F (S + s, U + u, w) when the anemometer rotor, by
means of a small electric motor, was brought into a steady rotation with a rate
independent of the wind speed in the tunnel. In this way they could determine the
function F in a range of the independent perturbation variables s, u and w. The
way the measurements were set up they were not able to vary the lateral velocity
component ṽ.
They found
   

 
 
 


 a1 
 
 1.03 < ±10% 


 
 
 


 
 
 


 
 
 


 
 
 ±0.1 



a2 
 

0.06 


 
 
 


 
 
 


 
 
 


 a3 
 
 −0.23 < ±10% 


 
 
 


 
 
 


 
 
 

 a4   0.96 < ±10% 
= . (15)

 
 
 


 a5 
 
 0.67 ±0.1 


 
 
 


 
 
 


 
 
 


 
 
 −0.73 < ±10% 


 a6 
 
 


 
 
 


 
 
 


 
 
 


 a7 
 
 0.04 ±0.1 


 
 
 


 
 
 


 
 
 

 a8   −0.12 ±0.1 
Applying (2), we may also write the second-order perturbation equation as
d s s u w
0 + = + µ1
dx S S U U
Λ s 2 u 2
− +
+Λ S +Λ U

1 v 2 µ21 µ2 w 2
+ + +
2 U +Λ 2 U
−Λ su 2µ1 u w
− +
+Λ S U +Λ U U
− Λ w s
− µ1 , (16)
+Λ U U
where we have assumed a calibration of the form (4).

Comparing (14) and (16), we see that there are the following constraints
a1 = 1, (17)
a3 + a4 + a6 = 0, (18)
a2 − a7 − a8 = 0 (19)
and
2a3 + a6 + 1 = 0. (20)
We see that the data (15) are consistent with these four relations† .
Coppin (1982) carried out a similar investigation on seven different cup anemome-
ters. Only a1 , a3 , a3 , a5 and a6 were determined in this investigation. The results
are shown in tabular form together with the result by Wyngaard et al. (1974).
Type a1 a3 a4 a5 a6
Friedrichs 1.27 0.00 1.10 0.85 −1.17
Siggelkow 1.18 −0.05 1.03 0.90 −0.90
Teledyne 3 1.25 −0.19 1.11 0.82 −0.86
Teledyne 6 1.26 −0.20 1.15 0.47 −0.89
Gill 3 1.22 −0.20 1.14 ≈0 −0.79
Casella 1.19 −0.29 0.98 ≈0 −0.69
Thies 1.18 −0.11 1.18 0.95 −1.26
Wyngaard 1.03 −0.23 0.96 0.67 −0.73
As the accuracy of the measurements is not stated explicitly in the 1982 investiga-
tion by Coppin it is not possible to judge if the relations (17), (18), (19) and (20)
are consistent with the measurements. It is claimed though that (18) is satisfied
within the experimental accuracy in all seven cases.
It is interesting to note that no positive values of a3 were found. We assume that
this is generally true for all cup anemometers and, since a3 = −Λ/( + Λ), the
implication is that Λ is never negative for a cup anemometer. This in turn means
that the equation F (S, U, 0) = 0 has only one root for which U/S = is positive‡ .
2.3 Mean Wind-Speed Measurements

The automatic recording of the cup anemometer signal can be done in many
ways. We must realize, however, that even if the wind speed is constant, the
angular velocity s̃ of the rotor varies over one full rotation. This means that the
rotation rate must be averaged over a least on full revolution before recording
and interpretation. If it takes the time ∆t for the cup rotor to complete one full
revolution, the average wind speed in this period is 2π/∆t. When determining
the average wind speed U over the period T we simply calculate N × 2π/T , where
N is the number of revolutions of the rotor in the period T . Roughly stated, this
corresponds to the displacement N × 2π of an air particle in the period of time
T . It is quite easy to build an electronic system which gives an electric pulse for
every rotor revolution and then it is just a matter of counting how many pulses
there are in the period T .
† Combining (19) and (20), we see that a (2a +a )+a +a = 0. In Kristensen (1993) there is
2 3 6 7 8
an unfortunate mistake in the corresponding equation (51) which reads 2a2 (a3 +a6 )+a7 +a8 = 0.
‡ The requirement that there could be only one steady-state calibration expression for a cup
anemometer was used by Kristensen (1993) as a proof that Λ must be positive. Of course this
argument is incorrect as the phenomenological model (2) is assumed valid and applied only close
to a steady state where (s̃, h̃) is close to (S, U ).
Let us for a moment consider what this means. The average wind speed Ui during
the ith revolution is 2π/∆ti , where ∆ti is the time it takes for the rotor to perform
the ith revolution. According to the definition of U , we have
N
−1
N 2π N 2π 1 1
U= = N = . (21)
T i=1 ∆ti
N i=1 Ui
We could imagine another procedure. We could determine, also by simple means,

the duration ∆ti of each revolution and then calculate the average of all values of
Ui = 2π/∆ti according to the equation
N
= 1
U Ui . (22)
N i=1
We note that
≥ U,
U (23)
where equality holds when all the Ui ’s are the same, i.e. if the wind speed is
constant. To see this, we study the ratio
N N N N

U 1 −1 1 Ui Uj
≡ 2 Ui U = + . (24)
U N i=1 j=1 j 2N 2 i=1 j=1 Uj Ui
Now, let
Ui
x= (25)
Uj
and
1
f (x) = x + . (26)
x
We see immediately by differentiation of f (x) that this function has one minimum
at 2 for x = 1, corresponding to Ui = Uj . Using this in (24), we conclude
N N

U 1
≥ 2 = 1. (27)
U 2N 2 i=1 j=1
Exactly by how much U is an overestimation depends primarily on how much Ui

varies with i. It can quite easily be shown that in the limit N → ∞ we have
≈U σU
U 1+ , (28)
U2
2
where σU is the variance of all the measured Ui ’s.
Later we will discuss how to operate a cup anemometer together with a wind vane
to obtain mean value and direction of the horizontal wind vector.
3 The Propeller Anemometer
Figure 3. Gill Propeller. The rotor diameter is 0.19 m. [From Busch et al. (1980)]
Figure 3 shows a Gill propeller. We consider a coordinate system with the mean
wind vector U in the direction of the propeller axis. The equation of motion is
then

s̃˙ = F (s̃, ũ, ṽ 2 + w̃2 ), (29)
where ũ, ṽ and w̃ are the instantaneous horizontal and vertical wind components
and s̃ is the instantaneous rotation rate of the anemometer rotor in rad s−1 . The
angular momentum of the rotor is proportional to s̃ and (29) just states that the
rate of change of the angular momentum is equal to the torque on the rotor. This
torque is caused by the wind and the friction in the instrument bearings.
We assume that the dynamics of the propeller can be described by the same equa-
tion as that pertaining to a cup anemometer and given by Kristensen (1993),
except that the dependence of the lateral wind component is the same as the de-
pendence of the vertical wind component, that the angular response is symmetric
and that there is no u-bias. This last requirement implies according to Kristensen
(1993) that Λ = ∞ in (2) so that
{ũ − s̃}s̃ ṽ 2 + w̃2

F (s̃, ũ, ṽ 2 + w̃2 ) = + µ2 . (30)
0 20
Here is the calibration distance, 0 the distance constant and µ2 a measure of

the sensitivity to the wind component perpendicular to the propeller axis. This
last quantity can be defined in terms the angular response function G(ϑ) (see Kris-
tensen (1993)), where ϑ is the angle between the wind velocity and the propeller
axis.
µ2 2
G(ϑ) = cos(ϑ) + ϑ (31)
2
and we note that µ2 = 0 corresponds to a cosine response.

When ũ is constant and equal to U while ṽ = w̃ = 0, we obtain after some time
s̃˙ = 0, i.e. s̃ becomes constant. Denoting this constant rotation rate S, the steady-
state calibration is obtained, as in the case of the cup anemometer, by solving
F (S, U, 0) = 0. (32)
The model (30) implies that S and U are proportional and related by
U = S. (33)
According to Busch et al. (1980) the calibration is linear. Here we also assume
that there is proportionality between U and S.
4 The Wind Vane

In order to avoid future confusion we will first specify the definition of the wind
direction.
According to the traditions in operational meteorology, the mean wind direction
D is defined as the direction from where the wind comes. In other words, the wind
direction is, according to this definition, the direction you are looking into, when
the wind is blowing in your face. If the wind blows from north then D = 0◦ and
D is measured clockwise so that east corresponds to D = 90◦ , south to D = 180◦
and west to D = 270◦ .
In physical meteorology we often work with a wind direction α̃ which is the direc-
tion of the mean wind velocity vector, i.e. opposite to D. It is measured counter-
clockwise and usually from east. The relation between D and the mean α̃ of α̃
is
D + α̃ = 270◦ (34)
or, if angles are measured in radians,
3
D + α̃ = π. (35)
2
In short term studies where the wind direction does not change systematically we
often chose to set the mean-wind direction equal to zero. We do that here to make
the discussion of wind-vane dynamics a little easier.
r φ̃˙
h̃
- o
˙
z −r φ̃
R
Iβ̃
oφ̃
6
α̃
z
U
Figure 4. Schematic illustration of the vane motion. Top view.
4.1 Wind-Vane Dynamics

Developing the equation of motion, we restate almost literally the derivation by
Larsen and Busch (1974). Figure 4 illustrates the motion of a wind vane. The
angle from the mean-wind velocity vector U with magnitude U to the horizontal
component of the instantaneous wind-velocity vector with magnitude h̃ is α̃. The
angles from this horizontal wind component and from U to the plane of the wind
vane are β̃ and φ̃, respectively. The vane is turning with the angular velocity φ̃˙ with
respect to the reference direction given by U . This means that the center of mass
of the vane is moving horizontally and perpendicular to the vane with the speed
˙
rφ̃, where r is the distance from the vane axis to the center of mass. Assuming
that all angles are small, it can be shown that the wind force is proportional to
the angle β̃ from the instantaneous horizontal wind-velocity component to the
moving vane. This angle is
˙
rφ̃
β̃ = β̃ + . (36)
U
The equation of motion (for angular momentum) of the vane is
1
I φ̃¨ = −r × ρU 2 AK β̃ , (37)
2
where I is the moment of inertia, A the vane area and K a dimensionless constant.
According to the theory for thin aerofoils (see e.g., Batchelor, 1967) K is 2π. Larsen
(1986) found K ≈ 1.9 for his vane. Here we just assume that K is of the order
unity.
Substituting (36) in (37) and using the relation
φ̃ = α̃ + β̃, (38)
we obtain the second-order differential equation for φ̃
¨ ˙
φ̃ + 2ζω0 φ̃ + ω02 φ̃ = ω02 α̃, (39)
where

ρrAK
ω0 = U (40)
2I
is a characteristic frequency and

r ρrAK
ζ= (41)
2 2I
a positive, dimensionless constant which is independent of U .

We note that
U
v ≡ (42)
ω0
is a characteristic length scale which does not depend on the wind speed. We will
call v the vane distance constant. Using (42), we may write
1 r
ζ= . (43)
2 v
Equation (39) describes a so-called damped oscillator with the natural frequency
ω0 and the damping coefficient ζ. The right-hand side, the instantaneous wind
direction, is the the input.
As in the case of a first-order, we can study the characteristics of this damped
oscillator by looking at its response to a step function.



 0 for t < 0
α̃ = . (44)


 α for t ≥ 0
The output from (39) is



 − − 2ω t

 1 exp(−ζω 0 t) cos 1 ζ 0







 + ζ 1 − ζ 2 ω0 t , 0≤ζ<1

 sin

 1 − ζ2


φ̃(t) = α 1 − exp(−ω0 t) {1 + ω0 t} , ζ=1 . (45)







 1 − exp(−ζω0 t) cosh ζ 2 − 1 ω0 t









 +
ζ
ζ − 1 ω0 t , 1<ζ<∞
 sinh 2
ζ2 − 1
ζ = 1/2
φ̃(t) ζ=1
ζ=2
0 3 6 9 12
ω0 t
Figure 5. Response of a damped oscillator to a step input.
Figure 5 shows how the response depends on the damping coefficient ζ. When
ζ ≥ 1 the solution (45) is an increasing function of time whereas it becomes a
damped oscillation for 0 < ζ <√1. In Figure 5, we we see the response (solid
line) φ̃(t) with the value ζ = 1/ 2. This damping coefficient provides the best
compromise between a fast response and minimal ‘overshoot’. The vane reacts in
this case as a so-called Butterworth filter.
In the limit ζ 1 (45) approaches

ω0 t
φ̃(t) = α 1 − exp − , (46)
2ζ
which is the response of a first-order filter with the time constant τ0 = 2ζ/ω0 .
According to (42) and (43) this time constant can be expressed as
r
τ0 = (47)
U
so that the effective vane distance constant eff simply becomes equal to U ×τ0 = r.
Figure (6) shows the difference between (45) and (46) when ζ = 2.
1
φ̃(t)
0.5
0
0 2 4 6 8 10 12 14
ω0 t
Figure 6. Response of a damped oscillator to a step input for ζ = 2 when φ̃(t) is
given by (45) (solid line) and by (46) (dotted line).
As discussed in subsection 2.2, the cup anemometer is a genuine first-order filter

with a distance constant 0 . If we want to operate the vane together with a cup
anemometer it would be best from a theoretical point of view that the vane can
also be considered a first-order filter with a distance constant eff = r = 0 . Since
0 is at least about 1 m for a conventional cup anemometer the vane arm r must
be rather long. Assuming that the vane is a light (styrofoam) aerofoil with height
h and width d, positioned at the end of the vane arm, the main contribution to
the moment of inertia I comes from this arm. Denoting the mass per unit length
of the arm σt , we have
1
I≈ σt r3 (48)
3
so that (41) becomes

ρhd
ζ≈B , (49)
σt
where B is a dimensionless constant of the order one.

We see that ζ is independent of r. The only ways in which we can increase damping
coefficient is therefore by decreasing σt of by increasing the area hd of the vane.
The direction sensor described by Larsen and Busch (1974) and by Larsen (1986)
has a styrofoam vane with the dimensions h = 20 cm, d = 10 cm and thickness
0.7 cm. The the arm consisted of two parallel steel tubes of length (= r + d/2)
20 cm and with σt = 0.2079 g cm−1 . Inserting in (49) and using ρ = 0.0013 g cm−3
for the density of air, we obtain ζ ∼ 1. It should be possible to construct a wind
vane with properties matching those of a cup anemometer.
In some situations as when studying atmospheric dispersion we are more interested
in determining the wind direction variance than in matching the filter character-
istics of a cup anemometer. Since the wind vane, like the cup anemometer, is by
itself a mechanical low-pass filter, a wind direction record will never be able to
show the full contribution to the variance from the high frequencies far beyond
U/v . However, since the wind-vane filter is of second order, we can take advantage
of the fact that an underdamped, second-order filter ‘overshoots’ at frequencies
close to U/v so that measured variance around U/v compensates for the high-
frequency loss. The next subsection is devoted to a discussion of how this might
be accomplished by designing a wind vane with a damping coefficient ζ close to
0.4.
4.2 Measuring the Lateral Velocity Variance

Let us start with the dynamic equation for the vane for small direction variations
in the deviation α ≡ α̃ − α̃ = v/U from the mean. Denoting the corresponding
vane response φ, we rewrite (39) as
v
φ̈ + 2ζω0 φ̇ + ω02 φ = ω02 . (50)
U
Since the distance constant v = U/ω0 is constant, it is more practical to operate

in the space domain by use of Taylor’s hypothesis. The equation of motion (50) is
therefore equivalent to
d2 φ 2ζ dφ φ 1 v
2 + dx + 2 = 2 U . (51)
dx v v v
The measured wind-direction variance φ2 is consequently given by
∞
Fv (k)dk
U 2 φ2 =
−∞ (1 − v k 2 )2 + 4ζ 2 2v k 2
2
∞ ∞
1
= Fv (k)dk + − 1 Fv (k)dk
−∞ −∞ (1 − 2v k 2 )2 + 4ζ 2 2v k 2
∞
2(1 − 2ζ 2 )2v k 2 − 4v k 4
= v 2 + 2 Fv (k)dk. (52)
0 (1 − 2v k 2 )2 + 4ζ 2 2v k 2
Assuming that the integral length scale for v is much larger than v , we may use
the inertial subrange expression
2
Fv (k) = α1 ε2/3 k −5/3 (53)
3
for the spectrum of v. For later purposes, we assume the values α1 k 4/3 = 0.165
and κ = 0.4.
The loss of variance can now be evaluated. The result is
4
v 2 − U 2 φ2 = α1 (εv )2/3 {h(ζ) − 2(1 − 2ζ 2 )g(ζ)}, (54)
3
where
∞
s1/3
g(ζ) = ds (55)
0 (1 − s2 )2 + 4ζ 2 s2
and
∞
s7/3
h(ζ) = ds. (56)
0 (1 −s2 )2
+ 4ζ 2 s2
These integrals can be expressed in terms of well-known functions as

 2 −1

 sin tan ( 1 − ζ /ζ)
2

 3

 , 0<ζ <1

 2ζ 1 − ζ 2


π
g(ζ) = √ × 1 , ζ =1 (57)
3 
 3

 2

 sinh ln(ζ + ζ − 1)
2



 3
, 1<ζ <∞
2ζ ζ 2 − 1
and

 4 −1

 sin tan ( 1 − ζ /ζ)2

 3

 , 0<ζ<1

 2ζ 1 − ζ 2


π
h(ζ) = √ × 2 , ζ=1 , (58)
3 
 3

 4

 sinh ln(ζ + ζ − 1)
2



 3
, 1<ζ<∞
2ζ ζ 2 − 1
which are shown in Figure 7.
10
1
h(ζ)
g(ζ)
0.1
0.1 1 10
ζ
Figure 7. The functions g(ζ) and h(ζ) given by (57) and (58), respectively.
Figure 8 displays the function
∆(ζ) = h(ζ) − 2(1 − 2ζ 2 )g(ζ). (59)
3
2
∆(ζ)
1
−1
−2
−3
−4
0.1 1
ζ
Figure 8. The function ∆(ζ) given by (59).
√
We see that when ζ is smaller than a certain value ζ0 < 1/ 2, then the loss of
variance is negative, i.e. that the measured variance is too large.
To determine ζ0 we must solve the equation
h(ζ0 ) = 2(1 − 2ζ02 )g(ζ0 ). (60)
Using (57) and (58), we obtain, by applying standard operations on trigonometric

functions, that this implies that

1
s tan tan−1 (s) = 1, (61)
3
where

1 − ζ02
s= . (62)
ζ0
After a slight re-arrangement (61) becomes

1
tan−1 (s) − 3 tan−1 =0 (63)
s
or

−1 1
s − tan 3 tan
s
= 0. (64)
1
1 + s tan 3 tan−1
s
Further,
3 − tan2 (u)
tan(3u) = , (65)
1 − 3 tan2 (u)
so that (64) implies
s2 − 3 = 3 − s−2 (66)
with the possible solutions
√
s2 = 3 ± 8. (67)
Since
1
ζ0 = √ , (68)
1 + s2
we must seek ζ0 among the solutions
1
ζ0 = √ . (69)
4± 8
√
We know already that ζ0 must be smaller than 1/ 2 and the only possibility is
then
1
ζ0 = √ ≈ 0.383, (70)
4+ 8
which, by insertion, is easily seen to be a solution.

For neutral stratification we have
v 2 = 2.68u2∗ (71)
and
u3∗
ε= , (72)
κz
so that the relative loss of variance becomes
2/3
v 2 − U 2 φ2 v
= 0.513 × ∆(ζ) × . (73)
v 2 z
v = 0.5 m is a typical value for a fast vane. When z = 10 m the loss of variance
becomes about −5% for ζ = 0.3 and 17% for ζ = 1.
Figure 9 illustrates the effect of the filtering when ζ = ζ0 : the area between the
two curves to the left of their intersection is equal to the area between the curves
to the right of the intersection.
It is very easy to extend the analysis to include spectra which follow a more general
power law, i.e.
gain kFv (k)
kFv (k) - (1−2v k 2 )2 +4ζ02 2v k 2
loss
0.1 1 10 100
v k
Figure 9. The filtered and unfiltered spectrum of the lateral velocity component in
an area conserving representation.
Fv (k) = Ak −β . (74)
If the integral in (52) is convergent, the constant β is limited to the range
1 < β < 3. (75)
The loss of variance is now
∞
s4−β − 2(1 − 2ζ 2 )s2−β
v − U φ =
2 2 2
2Aβ−1
v ds
0 (1 − s2 )2 + 4ζ 2 s2
= 2Aβ−1
v {hβ (ζ) − 2(1 − 2ζ 2 )gβ (ζ)} (76)
with
   
  
 sin((1 − −1 


 gβ (ζ) 
 
 β) cos (ζ)) 

π ζ 1 − ζ2
 
= π × −1 
0 < ζ < 1, (77)
  4 cos(β 2 ) 
 hβ (ζ)   sin((β −
3) cos (ζ)) 

 
ζ 1 − ζ2
where we have left out the case of ζ values ≥ 1.
Applying the identity
cos(2 cos−1 (ζ)) = 2 cos2 (cos−1 (ζ)) − 1 = 2ζ 2 − 1, (78)
we obtain
π sin((1 + β) cos−1 (ζ))

∆β (ζ) = hβ (ζ) − 2(1 − 2ζ 2 )gβ (ζ) = − π . (79)
4 cos(β ) ζ 1 − ζ2
2
It is very easy from (79) to determine the value ζ0 of the damping coefficient for
which the variance loss is exactly zero. Since this value is limited to the open
interval between zero and one, we get immediately

π
ζ0 = cos . (80)
β+1
We find several special values:
√
β = 23 =⇒ ζ0 = cos 2π cos(72◦ ) = 5−1 ≈
5 = 4 0.31

β = 35 =⇒ ζ0 = cos 3π
8 = cos(67.5◦ ) = 1
√ ≈ 0.38
4+ 8

β=2 =⇒ ζ0 = cos π
3 = cos(60◦ ) = 1 =
2 0.50
The filtered and unfiltered spectra are displayed for these three cases in Figures
10, 11 and 12.
2 kFv (k)
(1−2v k 2 )2 +4ζ02 2v k 2
kFv (k) -
1
0
0.1 1 10 100
Large ellv k
Figure 10. The filtered and unfiltered spectrum of the

√ lateral velocity component in
an area conserving representation. β = 3/2, ζ0 = ( 5 − 1)/4.
We may quantify the amount of lost variance which is recovered by using a filter
with a damping coefficient given by (80). One way is to determine the area between
the two curves in e.g. Figure 10 and normalize it by the true variance in the same
wave-number interval. This dimensionless number is a function only of β and is
given by

(β−1)
π
D(β) = (β − 1) 2 sin
2(β + 1)
 
∞  
1
× 1− s−β ds (81)
 2 2 2 π
(1 − s ) + 4 cos β+1 s 2 
π
2 sin( 2(β+1) )
4
2 kFv (k)
kFv (k) - (1−2v k 2 )2 +4ζ02 2v k 2
0
0.1 1 10 100
Large ellv k
Figure 11. The filtered and unfiltered spectrum of the

lateral√velocity component in
an area conserving representation. β = 5/3, ζ0 = 1/ 4 + 8.
2 kFv (k)
kFv (k) - (1−2v k 2 )2 +4ζ02 2v k 2
0
0.1 1 10 100
Large ellv k
Figure 12. The filtered and unfiltered spectrum of the lateral velocity component in
an area conserving representation. β = 2, ζ0 = 1/2.
and displayed in Figure 13.

It is easy to evaluate D(β) in the limit β → 1. We get
lim D(β) = 1, (82)

β→1
as also indicated by Figure 13.
1
0.9
D(β)
0.8
0.7
0.6
0.5
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
β
Figure 13. The function D(β) given by (81).
5 Cup and Vane

We have seen that the mean wind speed was given by (21). This equation shows
that U T is the total distance a hypothetical air particle would travel in the period
T if the wind were constant in space (but not necessarily in time). Each revolution
of the cup rotor corresponds to the passage through the anemometer of a column
of air of length 2π.
However, the average wind-vane direction φ̃i will in general change from one rev-
olution to the next and, if we want to know how far from the starting point our
hypothetical air particle has traveled horizontally in the period T , we must add all
the small distances 2π vectorwise. This procedure is illustrated in Figure 14. We
will redefine the average wind speed as the length of the resulting vector divided
by T and wind direction as the direction of this vector. With the x-axis pointing
east and the y-axis pointing north, the resulting displacement vector has the two
components (U T, V T ).
In mathematical terms (U, V ) are given by
 N 
   
  
 

 U 
  
 cos(φ̃i ) 

2π i=1
= N . (83)

  T  
 V 
 

 sin(φ̃i )



 
i=1
The mean direction φ̃ is then computed from the equation

V
φ̃ = arctan . (84)
U
The practical way of obtaining the information about both speed and direction
is to let each full rotation of a cup anemometer trigger a recording of cos(φ̃) and
N
6
N × 2π
VT >
2π
1
7
-
I
φ̃
*
]φ̃1 - E
UT
Figure 14. Illustration of the vector wind run.
sin(φ̃). It is important in this context that the distance constant of the anemometer
0 matches the distance constant of the wind vane eff .
Kristensen (1993) pointed out that, as a bonus, also the (filtered) wind direction
variance can be obtained using this type of recording. By expanding cos(φ) 2 and
sin(φ) 2 in the deviation δφ from φ , a simple analysis shows
δφ2 = 1 − cos(φ) 2 − sin(φ) 2 + terms of order δφ2 2 and higher. (85)
In other words, due to the linearity of the cup anemometer calibration it is possible
in a simple way to obtain three mean quantities, wind speed, wind direction and
wind-direction variance—quantities important in e.g. dispersion measurements.
6 Prop and Vane

Figure 15 is a schematic diagram of the propeller-vane anemometer.
When the propeller is guided into the wind by means of the vane we realize that
the propeller axis cannot be aligned exactly along the mean wind
r φ̇
K
U j0
- j
K 6 i
* ]φ
◦*
-
i0
a

Figure 15. Schematic top view of the propeller-vane geometry and motion.
U = ũ = ũ i0 , (86)
where
ũ ≡ ũi0 + ṽj 0 + w̃k0 ≡ h̃ + w̃k0 . (87)
The mean wind direction is characterized by the unit vector i0 , the lateral and
the vertical directions by the unit vectors j 0 and k0 .
The three components of the instantaneous wind speed are
   

  
  


 ũ 
 
 U +u 


 
 
 


  
  

 ṽ  =  v 
, (88)

 
 
 


 
 
 


   
 
 w̃   w 

where
u = v = w = 0. (89)
The instantaneous angle from the mean wind direction i0 to the direction of the
vane axis i is φ. The center of the propeller has the distance a from the vane axis
so that the position of this center is −a cos(φ) i0 − a sin(φ) j 0 . Consequently, the
velocity of the propeller center becomes
d
vp = {−a cos(φ) i0 − a sin(φ) j 0 } = aφ̇ sin(φ) i0 − aφ̇ cos(φ) j 0 . (90)
dt
The instantaneous wind velocity, experienced by the propeller in the moving ref-
erence frame of the vane, is therefore
ũ − v p = {U + u − aφ̇ sin φ} i0 + {v + aφ̇ cos φ} j 0 + w k0 . (91)
The relation between the coordinate system i0 , j 0 , k0 , determined by the mean

wind vector, and the coordinate system of the vane i, j, k is
    

 
 
 
 
 


 i0 
 
 cos(φ) − sin(φ) 
0  
 i 


 
 
 
 
 


 
 
 
 

 j0 
=
 sin(φ) cos(φ) 0  j , (92)

 
 
 
 
 


 
 
 
  

     

 k0 
 
 0 0 1  k 
  
so that the relative velocity ũ − v p in the propeller coordinate system becomes
ũ − v p = {U + u − aφ̇ sin(φ)}{cos(φ) i − sin(φ) j}
+ {v + aφ̇ cos(φ)}{sin(φ) i + cos(φ) j} + w k
= {U cos(φ) + u cos(φ) + v sin(φ)} i
− {U sin(φ) + u sin(φ) − v cos(φ) − aφ̇} j + w k. (93)
We assume that φ 1 and that the numerical values of u, v and w are much
smaller than U . Further, we assume that a|φ̇| U and expand (93) to second
order in the small quantities. The result is
ũ − v p ≈ U i + u i + {v + aφ̇ − U φ} j + w k
U 2
+ {vφ − φ } i − uφ j. (94)
2
We substitute (94) and
s̃ = S + s (95)
in (29) and, using the expression (30), we obtain to second order
0 u us − s2
ṡ + s ≈ +
U U
vφ − 12 U φ2 {v + aφ̇ − U φ}2 + w2
+ + µ2 . (96)
2U
Equation (96) is a first-order ordinary differential equation in s. The right-hand
side contains first and second-order terms. The known variables are u, v and w.
The fourth variable φ is determined by the dynamic equation (50) for the vane.
We have defined the velocity variables such that they fulfil (89). Equation (50)
implies that also
φ = 0. (97)
However, the way s is defined by (95) does not guarantee the mean of s is zero.
In fact, taking the average of (96), we get
1
s = {us − s2 }
U
1 U µ2
+ {vφ − φ2 } + {(v + aφ̇ − U φ)2 + w2 }. (98)
2 2U
In order to determine s , we must evaluate all the second-order terms on the

right-hand side of (98). The terms involving s are zero in our approximation since
s to the first order fulfills the differential equation
0 u
ṡ + s = . (99)
U
Multiplying this equation on both sides by s and averaging yield
s2 = us /, (100)
which shows that the two first term on the right-hand side of (98) cancel.
In order to determine the other terms we must solve (50) with the initial conditions
φ(−∞) = 0 and φ̇(−∞) = 0. The result is
∞
ω0 1
φ(t) = v(t − τ )e−ζω0 τ sin( 1 − ζ 2 ω0 τ )dτ. (101)
1 − ζ2 U 0
We obtain immediately
∞
1 ω04 Sv (ω)
φ2 = dω (102)
U2 2
−∞ (ω0 − ω 2 )2 + 4ζ 2 ω02 ω 2
and
∞
1 ω02 (ω02 − ω 2 )Sv (ω)
vφ = dω, (103)
U −∞ (ω02 − ω 2 )2 + 4ζ 2 ω02 ω 2
where Sv (ω) is the power spectrum of v.

Multiplying (50) by either φ or by φ̇, followed by averaging, we easily obtain the
two relations
vφ
−φ̇2 + ω02 φ2 = ω02 (104)
U
and
v φ̇
2ζω0 φ̇2 = ω02 , (105)
U
so that
φ̇2 v φ̇ vφ
2 = = φ2 − . (106)
ω0 2ζω0 U U
With (106) we can now evaluate the relative bias δ = s /S in terms of φ2 and
vφ /U :

vφ φ2
δ = (1 − µ2 ) −
U 2
2
v + w2 a2 ω02 aω0 vφ

+ µ2 + + 2ζ φ −
2
. (107)
2U 2 2U 2 U U
Using (102) and (103), we get
v 2 + µ2 w2
δ =
2U 2

1 − µ2 ∞ ω 2 (ω 2 + 4ζ 2 ω02 )Sv (ω)
− dω
2U 2 −∞ (ω02 − ω 2 )2 + 4ζ 2 ω02 ω 2
2 2 ∞
µ2 a ω0 aω0 ω02 ω 2 Sv (ω)
+ + 4ζ dω. (108)
−∞ (ω0 − ω ) + 4ζ ω0 ω
2U 2 U 2 U 2 2 2 2 2 2
Introducing the vane distance constant v and applying, once again, Taylor’s hy-
pothesis, we can write (108) in the form
v 2 + µ2 w2
δ =
2U 2

1 − µ2 ∞ k 2 2v (k 2 2v + 4ζ 2 )Fv (k)
− dk
2U 2 −∞ (1 − k 2 2v )2 + 4ζ 2 k 2 2v
2 ∞
µ2 a a k 2 2v Fv (k)
+ + 4ζ dk. (109)
−∞ (1 − k v ) + 4ζ k v
2U 2 2
v v 2 2 2 2 2 2
Assuming that v is much smaller than the integral scale of v(t), we may use (53)
for Fv (k).
This leads to the expression
v 2 + µ2 w2
δ =
2U 2
1 − µ2 4
− α1 (εv )2/3 {4ζ 2 g(ζ) + h(ζ)}
2U 2 3
2
µ2 a a 4
+ + 4ζ α1 (εv )2/3 g(ζ), (110)
2U 2 2v v 3
where g(ζ) and h(ζ) are given by (57) and (58).
In the neutrally stratified surface layer we have (e.g. see Kristensen et al. (1989))
v 2 = 2.68 u2∗, (111)
w2 = 1.46 u2∗ (112)
and, assuming local balance between shear production and destruction by dissi-
pation of turbulent kinetic energy,
u3∗
ε= . (113)
κz
Using the same values as Kristensen et al. (1989) with α1 κ4/3 = 0.165 and κ = 0.4,
we can write (110) in the form
δ
= 1 + 0.54µ2
v 2 /(2U 2 )
2/3
v
− 0.51(1 − µ2 ){4ζ g(ζ) + h(ζ)}
2
z

2/3
a2 a v
+ 0.51µ2 2
+ 4ζ g(ζ) . (114)
v v z
As an example, let us apply (114) to the Gill propeller vane type 35003D. Accord-
ing to Monna (1978) and Monna and Driedonks (1979), this model has a ≈ 0.35 m,
v ≈ 1.0 m and ζ ≈ 0.6. Further, Monna and Driedonks (1979) found that the
angular response of the Gill propeller can be written
G(ϑ) = cos(1.3ϑ) ≈ cos(ϑ) − 0.35ϑ2 , (115)
which, by comparison with (31), shows that µ2 ≈ −0.7.

Inserting in (114) we obtain
2/3
δ v
= 0.62 − 3.3 (116)
v 2 /(2U 2 ) z
for this Gill propeller vane.

A general comparison between the propeller vane and a cup vane is most easily
carried out by going back to (110). The two first terms in this equation are the same
as those we found for the cup anemometer (Kristensen, 1993). They are the v-bias
and the w-bias, respectively. The rest of the propeller-vane bias in (110) we denote
δpv . This bias occurs partly because the vane is lagging behind the instantaneous
wind direction and partly because the vane itself moves with respect to the wind.
We see that the two terms in δpv have a common dimensionless factor, namely
2 (εv )2/3
B= α1 . (117)
3 U2
In the horizontally homogeneous, diabatic surface layer both U and ε are stability
dependent so in general B is also stability dependent.
Following e.g. Panofsky and Dutton (1984), we have
u3∗ z
ε= ϕε (118)
κz L
and
z
u∗ z
U= ln − ψm , (119)
κ z0 L
where L is the Monin-Obukhov length and z0 the roughness length.

The two functions ϕε (z/L) and ψm (z/L) are
z z z
ϕε = ϕm − (120)
L L L
and
z z/L z/L
ds ds
ψm = [1 − ϕm (s)] ≈ [1 − ϕm (s)] , (121)
L z0 /L s 0 s
where
z κz dU
ϕm ≡ . (122)
L u∗ dz
According to Panofsky and Dutton (1984) we have



 (1 − 16s)−1/3 , s≤0
ϕm (s) = , (123)


 1 + 5s , s>0
so that for z/L ≤ 0
z
3 1 + ξ + ξ2 √ −1 1 ξ−1
ψm = ln − 3 tan √ (124)
L 2 3 3ξ+1
with ξ = 1/ϕm (z/L), while for z/L > 0
z z
ψm = −5 . (125)
L L
We can now write
2/3
z z v
B = χpv , × , (126)
z0 L z
where
z
−2 z
z z 2 z
χpv , = φ1κ
4/3
ln − ψm ϕ2/3
ε . (127)
z0 L 3 z0 L L
Finally, we obtain
2
a a
δpv = − (1 − µ2 )(4ζ 2 g(ζ) + h(ζ)) − µ2 + 4ζ g(ζ)
2v v
2/3
z z v
×χpv , × . (128)
z0 L z
The function χpv (z/z0, z/L) is displayed in Fig. 16 for two typical values of z/z0 .
0.1

χpv ,
z z
z
z0 = 100
z0 L
0.01
z
z0 = 1000
0.001
−5 −4 −3 −2 −1 0 1 2
z/L
Figure 16. The function χpv (z/z0 , z/L) as given by (127).
7 Conclusions
This report has been inspired by a number of unanswered questions which arose
during and after the writing of the report by Kristensen (1993). Consequently, it
has not been possible to keep as well a logical order in dealing with these questions.
Two of the errors which have later been found in Kristensen (1993) are reported
here in footnotes.
Concerning the cup anemometer, the question of overspeeding was discussed in
much detail by Kristensen (1993) so we have only restated, in section 2 the phe-
nomenological model for the forcing of the cup rotor and pointed out that there
is no inconsistencies between this model and the experimental findings by Wyn-
gaard et al. (1974) and Coppin (1982). Further is pointed out in subsection 2.3
that simple determination of the mean-wind speed in a sense is ambiguous since
of the mean-wind speeds in each revolution of
a calculation of the time average U
the cup rotor is systematically larger than the mean wind speed U calculated by
counting the number of rotor revolutions during the averaging period. The overes-
timation is approximately proportional to the ratio of variance of the mean wind
speeds in each revolution and U 2 and thus more important the more turbulent
the wind.
In section 3 we apply the same type of phenomenological model to the forcing of the
propeller. In a way this model is simpler than that for the cup anemometer since it
requires only three parameters, the calibration distance , the distance constant 0
for a flow along the propeller axis, and a dimensionless parameter µ2 specifying the
symmetric angular response. The corresponding model for the cup anemometer
has two more parameters, namely one Λ of dimension length and a dimensionless
parameter µ1 specifying the asymmetric part of the angular response. There has
been no experimental verification of the model for the propeller model.
Section 4 is devoted entirely to the wind vane. The equation of motion which is
derived according to Larsen and Busch (1974) is, in contrast to the first-order
differential equations for the cup and for the propeller, of second-order. Therefore
the filter characteristics must be described not only by a distance constant v , but
also by a damping coefficient ζ. When operated together with a cup or propeller
anemometer, it is important to match the distance constants and try to design the
vane such that the vane is so overdamped that it approximates a first-order filter.
It is argued that, since ζ for sufficiently light vane material is only a function of
the vane area and the mass per unit length of the vane arm, it should be possible
to match the characteristics of a cup anemometer. This remains to be shown ex-
perimentally. In subsection 4.2 we take another approach and show that an under-
damped wind vane can be used to determine the wind direction variance without
loss. It turns out that if v is much smaller than the scale of the turbulence then—
theoretically—the high-frequency loss of measured variance can be compensated
by the ‘overshooting’, at lower frequencies, of the underdamped vane. √
This com-
pensation balances the loss almost exactly if ζ = cos(3π/8) = 1/ 4 + 8 ≈ 0.38.
It would be very interesting if this could be verified experimentally as it would
be useful to measure the lateral variance of the turbulent wind in e.g. studies of
turbulent atmospheric dispersion. Actually, it should be easier to construct an
underdamped than an overdamped wind vane.
In section 5 we restate an important result from Kristensen (1993). We show that
the calibration (4), i.e. S = U/, where the calibration is an instrument constant,
suggests that the best way to obtain the mean wind magnitude and direction is
to measure, from each revolution of the cup rotor, sine and cosine of the wind
direction and average both of these quantities over the averaging time T . In this
way we obtain a magnitude which is equal to the so-called wind way, or vector
wind run, divided by T (see Figure 14) and, at the same time get an estimate of
the wind-direction variance.
Finally, in section 6 we discuss the propeller-vane anemometer. The main ques-
tion here is whether this instrument has systematic errors like those of the cup
anemometer (Kristensen, 1993). It turns out that what is called v-bias and w-bias
of a cup anemometer is of exactly the same form for a propeller-vane anemome-
ter. In addition, there are two other sources of bias on the mean-wind speed. The
first can be understood qualitatively by noting that the propeller is always lacking
behind and will never has its axis aligned with the instantaneous horizontal wind
component. Since the wind forcing on the propeller is always, for a given magni-
tude of the speed, at maximum when the wind is directed along the propeller axis,
the rotation rate will always be systematically too small. Therefore this bias con-
tribution is negative. The other source can also be interpreted qualitatively: The
vane is rotating in the horizontal plane around a vertical axis and is displaced the
distance a with respect to this axis (see Figure 15). Therefore the propeller has its
own motion with respect to the air and, even if the wind speed were exactly zero,
the propeller would turn if the vane were turning. The sign of this bias depends
on the angular-response parameter µ2 and the ratio of a to the distance constant
v of the vane. We note that the propeller distance constant 0 , with the model
postulated here, does not have any influence on the bias.
References
Batchelor, G.K. (1967). An Introduction to Fluid Dynamics, Cambridge University
Press, 615 pp.
Brazier, C.E. (1914). Recherches expérimentales sur les moulenets anémometri-
que. Ann. Bur. Centr. Météorol. France, 157-300; Thèse Fac. Sc. Univ. Paris
(1921), ed. Gaurthier-Villars.
Busch, N.E., Christensen, O., Kristensen, L., Lading, L. and Larsen, S.E. (1980).
Cups, Vanes, Propellers and Laser Anemometers, Air-Sea Interaction —
Instruments and Methods., F. Dobson, L. Hasse and R. Davis, Eds., Plenum
Press, NY, 11-46.
Coppin, P.A. (1982). An Examination of Cup Anemometer Overspeeding. Mete-
orol. Rdsch., 35, 1-11.
Kristensen, L., Lenschow, D.H., Kirkegaard, P. and Courtney, M.S. (1989).
The Spectral Velocity Tensor for Homogeneous Boundary-Layer Turbulence,
Boundary-Layer Meteorol., 47, 149-193.
Kristensen, L. (1993). The Cup Anemometer and Other Exciting Instruments,
Risø-R-615(EN), Risø National Laboratory, Roskilde, 83 pp.
Larsen, S.E. and Busch, N.E. (1974). Hot-wire Measurements in the Atmosphere.
Part I: Calibration and Response Characteristics. DISA Information, 16, 15-
34.
Larsen, S.E. (1986). Hot-wire Measurements of Atmospheric Turbulence Near the
Ground, Risø-R-233, Risø National Laboratory, Roskilde, 342 pp.
Middleton, W.E.K. (1969). Invention of the Meteorological Instruments, The Johns
Hopkins Press, Baltimore, MD, 362 pp
Monna, W.A.A. (1978). Comparative Investigation of Dynamic Properties of some
Propeller Vanes, W.R. 78-11, K.N.M.I., de Bilt, the Netherlands, 30 pp.
Monna, W.A.A. and Driedonks, A.G.M. (1979). Experimental Data on the Dy-
namic Properties of Several Propeller Vanes, J. Appl. Meteor., 18, 699-702.
Panofsky, H.A. and Dutton, J.A. (1984). Atmospheric Turbulence, John Wiley &
Sons, NY, 397 pp.
Patterson, J. (1926). The Cup Anemometer. Trans. Roy. Soc. Canada, Ser. III,
20, 1-54.
Wyngaard, J.C., Bauman, J.T. and Lynch, R.A. (1974). Cup Anemometer Dy-
namics. Proc. Instrument Society of America, Pittsburgh, PA, May 10-14,
1971, 1, 701-708.
Wyngaard, J.C. (1981). Cup, Propeller, Vane, and Sonic Anemometers in Turbu-
lence Research. Annu. Rev. Fluid Mech., 13, 399-423.
Bibliographic Data Sheet Risø–R–766(EN)
Title and author(s)
Cups, Props and Vanes
Leif Kristensen
ISBN ISSN
87–550–2006–2 0106–2840
Dept. or group Date
Meteorology and Wind Energy August 1994

Groups own reg. number(s) Project/contract No.
Pages Tables Illustrations References
37 1 16 15
Abstract (Max. 2000 char.)
We discuss the dynamics of the cup anemometer, the propeller anemometer and
the wind vane. The phenomenological model by Kristensen (1993) is modified to
describe the motion of the propeller anemometer. We use the well-know second-
order differential equation describing the motion of the wind vane to study the
properties of the vane itself and the vane in combination with a cup and a vane.
It is argued that even the simplest question about how to record the signal from
a cup anemometer has an ambiguous answer and can lead to a, not necessarily
small, systematic error in the measured mean wind speed. We show how it is
possible to measure the entire wind direction variance by utilizing the fact that
an underdamped vane by ‘overshooting’ may compensate for the high-frequency
variance loss if the damping coefficient is about 0.4. Finally we discuss possible
sources of bias on the measured mean-wind speed when using a propeller-vane
anemometer. It turns out that this anemometer has the same type biases from
the lateral and vertical wind fluctuations as has the cup anemometer. In addition
there are bias contributions from misalignment between the mean wind and the
propeller axis and from the translatory motion of the propeller itself with respect
to the vertical wind-vane axis.
Descriptors INIS/EDB
ANEMOMETER; DYNAMICS; FLUCTUATIONS; TURBULENCE; VANES;

VELOCITY; WIND
Available on request from Information Service Department, Risø National Laboratory

(Afdelingen for Informationsservice, Forskningscenter Risø), P.O. Box 49, DK–4000 Roskilde, Den-
mark.
Phone +45 46 77 40 04, Fax +45 46 77 40 13, E-mail infserv@risoe.dk

Cups Props and Vanes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cups Props and Vanes

Uploaded by

Copyright:

Available Formats

Risø–R–766(EN)

Cups, Props and Vanes

Risø National Laboratory, Roskilde, Denmark

Information Service Department · Risø · 1994

2 The Cup Anemometer 5

3 The Propeller Anemometer 13

4 The Wind Vane 14

5 Cup and Vane 26

6 Prop and Vane 27

2 The Cup Anemometer

2.1 Phenomenological Model

(h̃ + µ1 w̃ − s̃)(h̃ + µ1 w̃ + Λs̃) µ2 w̃2

ũ2 + ṽ 2 + w̃2 − 2 s̃2

where C is a dimensionless constant of order unity, ρ ≈ 0.0013 g cm−3 the density

2.2 Anemometer Dynamics

with w(t) = w̃ kept equal to zero.

—the classical response of a ﬁrst-order system to a step function. Figure 2 shows

in which case (8) takes the form

Figure 2. First-order response (11) of a cup anemometer to a step input ∆U .

Wyngaard et al. (1974) now determined experimentally the eight coeﬃcients

Applying (2), we may also write the second-order perturbation equation as

where we have assumed a calibration of the form (4).

Friedrichs 1.27 0.00 1.10 0.85 −1.17

Siggelkow 1.18 −0.05 1.03 0.90 −0.90

Teledyne 3 1.25 −0.19 1.11 0.82 −0.86

Teledyne 6 1.26 −0.20 1.15 0.47 −0.89

Gill 3 1.22 −0.20 1.14 ≈0 −0.79

Casella 1.19 −0.29 0.98 ≈0 −0.69

Thies 1.18 −0.11 1.18 0.95 −1.26

Wyngaard 1.03 −0.23 0.96 0.67 −0.73

2.3 Mean Wind-Speed Measurements

We could imagine another procedure. We could determine, also by simple means,

Exactly by how much U  is an overestimation depends primarily on how much Ui

{ũ − s̃}s̃ ṽ 2 + w̃2

Here is the calibration distance, 0 the distance constant and µ2 a measure of

and we note that µ2 = 0 corresponds to a cosine response.

4 The Wind Vane

D + α̃ = 270◦ (34)

or, if angles are measured in radians,

Figure 4. Schematic illustration of the vane motion. Top view.

4.1 Wind-Vane Dynamics

The equation of motion (for angular momentum) of the vane is

we obtain the second-order diﬀerential equation for φ̃

is a characteristic frequency and

a positive, dimensionless constant which is independent of U .

The output from (39) is

Figure 5. Response of a damped oscillator to a step input.

As discussed in subsection 2.2, the cup anemometer is a genuine ﬁrst-order ﬁlter

so that (41) becomes

where B is a dimensionless constant of the order one.

4.2 Measuring the Lateral Velocity Variance

Since the distance constant v = U/ω0 is constant, it is more practical to operate

The measured wind-direction variance φ2 is consequently given by

These integrals can be expressed in terms of well-known functions as

which are shown in Figure 7.

Figure 8 displays the function

∆(ζ) = h(ζ) − 2(1 − 2ζ 2 )g(ζ). (59)

Figure 8. The function ∆(ζ) given by (59).

h(ζ0 ) = 2(1 − 2ζ02 )g(ζ0 ). (60)

Using (57) and (58), we obtain, by applying standard operations on trigonometric

After a slight re-arrangement (61) becomes

so that (64) implies

with the possible solutions

we must seek ζ0 among the solutions

Exactly by how much U is an overestimation depends primarily on how much Ui

D + α̃ = 270◦ (34)

The measured wind-direction variance φ2 is consequently given by

The mean direction φ̃ is then computed from the equation

δφ2 = 1 − cos(φ) 2 − sin(φ) 2 + terms of order δφ2 2 and higher. (85)

U = ũ = ũ i0 , (86)

In order to determine s , we must evaluate all the second-order terms on the

s2 = us /, (100)

v + w2 a2 ω02 aω0 vφ

v 2 = 2.68 u2∗, (111)

w2 = 1.46 u2∗ (112)